Deploying Geospatial AI Models for Impact Evaluation

In this project we are adapting existing geospatial foundation models for use in impact evaluation settings, including generating dependent variables and adjusting for confounders. The substantive focus of the project will be to link satellite imagery with Kenyan census data and train a geospatial foundation model (see Clay) to predict outcomes like population density and wealth. We will use these predictions to estimate the impacts of large-scale resettlement schemes in Kenya.

Students working on this project will develop skills working with state of the art AI models for satellite imagery and will engage deeply with theoretical literature in causal machine learning. Preference will be given to students who expect to work on the project for a calendar year or more.